I don't see why we shouldn't do this. I think that [], (), (;) and Dict() can be safely considered as leafs

It would be pretty breaking, because we use exclude = isleaf to decide what to pass to map function callbacks. Code that never expected to encounter an empty (non-numeric) array or dict before is now going to have to deal with them. At a higher level, making two possible paths for particular types based on their contents over just types blows up any chance of us making traversals type stable and introduces additional overhead for users (who will now have to detect and guard against random empty arrays/dicts which they're not interested in. Distinguishing those from parameters would not be fun!)

As I mentioned in #52 (comment), the safer path seems to be limiting and not broadening the scope of isleaf so that we have fewer priveleged types.

right, I clearly didn't think it through.